Advanced Stats Matchup: Michigan vs. Michigan State

Submitted by Ecky Pting on

Advanced Stats Matchup Analysis

- 2017 Michigan vs. Michigan State

Introduction

Behold, another installment of the new and improved Advanced Stats (S&P+) Matchup, this time featuring Michigan vs. Michigan State.

This matchup analysis draws upon the Advanced Stats Profiles published weekly by Bill Connelly on Football Study Hall. The profiles feature Connelly’s well-known Five Factors, and also include the more detailed groups of S&P+ metrics that break down elements of the game such as Rushing and Passing, as well as the down-and-distance scenarios known as Standard Downs and Passing Downs. As you may recall from last season, this matchup analysis was presented in the form a somewhat lengthy table listing the 26 metrics (this season it’s only 20), with a column of metrics for the offense and defense for each team. Derived from these two pairs of metrics were two more columns of matchup metrics, which when compared would indicate which team held a net advantage in that metric. It was a lot to digest, and in the end, it failed to really provide a qualitative characteristic of how great (or negligible) an advantage was that a team held in any metric relative to the other metrics. To that end, this new approach seeks to display the matchups graphically, in a way that more clearly distinguishes and gauges the significance of any net advantages. For more details regarding the definition of and concepts behind each of the metrics, the Advanced Stats Glossary is a handy reference to bookmark.

Methodology

This section describes the approach to analyzing Bill Connelly’s base metrics, the formulation for deriving the matchup metrics and the format for the charts. None of this is etched in stone, and certainly suggestions for improving any of the aspects of the methodology are welcome and appreciated!

Technical Approach

The analysis evaluates metrics that are applicable to both offensive and defensive units of two competing teams, such that a set for a given metric consists of five values: Team A Offense, Team A Defense, Team B Offense and Team B Defense, and the National Average. From this set, two matchup values are derived. The first matchup value is “Team A Offense vs. Team B Defense,"  which as it states gauges the competitive performance of the Team A offense against the Team B defense. The resulting matchup value is then normalized to a matchup between two average teams, so a relative comparison can be made with the opposing team’s result, as well as with matchups for other metrics.

Formulations

The first matchup value is determined by simply taking the product of the Team A offense and Team B defense metrics, divided by the national average for the given metric. The second matchup value is in turn computed in the same way for the Team B Offense versus the Team A Defense. Once the two matchup metrics are determined, the team with the higher value when on offense will have a net advantage for that metric, with the exception of three categories: "Stuff Rate", "Standard Down Sack Rate" and "Passing Down Sack Rate". These are termed contra-metrics for the purposes of this diary. A contra-metric gauges the offense's ability to avoid the given categorical description. Akin to a contra-asset in the accounting vernacular, with these metrics, a lower value is better.

The one factor or metric that does not conform to this principle of geometric scalability described above - because it is predominantly a random variable with a zero mean - is Turnover Luck. The principle of Regression to the Mean would suggest that a team that accrues a negative TO Luck metric coming into a game would likely have better luck than in past games, and likewise the converse holds true for a team with a positive TO Luck metric. Thus, the team with a lower TO Luck metric could be expected to benefit the most in the ensuing game, and the relative benefit would amount to the difference in the TO Luck metrics between the two teams.

 

Data Visualization

The charts are arranged according to the groupings in the table above. All of the base metric numerical data, as well as matchup values, are embedded in the individual metric charts in the small table at the bottom. Metrics for each team’s Offense, Defense, and its Offense versus the opponent’s Defense are read across the designated row in the table. The same data is also depicted visually in chart graphic. Along each side is a vertical line plotted between two color-coded and shape-coded markers. The vertical line on the left side is for “Team A Off. vs. Team B Def.”, where the circle (or “O”) marker designates the value for the offense. Likewise, the diamond marker designates the value for the defense. The markers are in turn color coded according to the particular team’s colors. So you will notice that the color-coding is consistently reversed between the left and right sides across the charts. A third, horizontal dash marker designates the value for the composite matchup between the given offensive and defensive values, as determined by the formulation noted above.

Next is the block in the center of the graphic. The block simply gives emphasis to the vertical difference between the matchup values (the horizontal dash markers) on the left and right vertical lines. The blocks are in turn color coded according to the team whose offense corresponding to the greater matchup value.

Also included in each chart is a horizontal dashed line showing the FBS National Average value for each metric. This is the value to which the matchup values have been normalized.

Last - and this is where the rubber meets the road in setting up this visualization approach – is the Y-axis scaling across the charts (ignoring Turnover Luck, for which this does not apply). You may notice that a logarithmic scale has been applied, and this is because its better suited to reflect the geometric normalization that’s in play here (e.g. 2 times the average will have the same vertical offset as 1/2 of the average, just in the opposite direction). So what’s going on here is that the bounds of the vertical scale for each chart are set to the same multiple of the FBS National Average of each particular metric. For example, the maximum values for the first four charts are set to 6 times the FBS Average values. Likewise, the minimum values are pegged to 1/6 of the FBS Average values, so in the end, the plus or minus percentage range is the same for each chart relative to the FBS average for that chart.

From there, you can just eyeball the blocks, and easily observe which team has the advantage in which matchups, and evaluate whether the matchups are relatively close, as well as where there is potential for a mismatch.

Michigan vs. Michigan State Matchup Analysis

So, on with the matchup analysis!

The Five Factors Matchups

Here are the matchups for the core Five Factors metrics that compose the actual S&P+ ratings from which the game scoring margin is derived. As of the beginning of this week, that margin stands in favor of Michigan, to the tune of –14.7 points. Keep in mind a couple of things: the weightings of the factors into the predicted scoring margin are not uniform and, a team has control of only the first four. Of those first four, UM has an advantage in three, so there’s that, with a little Turnover Luck gravy ladled on top.

Efficiency

In Efficiency, the UM Offense is below average, while the MSU Defense is well above average, which pulls down the UM Offense. On the other side, the MSU Offense is slightly above average, however, the UM Defense is elite. In fact, it is Ranked #1 in this category! The net matchup gives a considerable edge to Michigan in Efficiency.

Explosiveness

In Explosiveness, the UM Offense is well above average, but the MSU Defense is also better than average, which pulls the UM Offense down to about average. On the other side, both the MSU Offense and UM Defense are below average, resulting in a slight improvement for the MSU Offense, but leaving it still below average. The net matchup gives just a slight edge for Michigan in Explosiveness.

Field Position

As for Field Position, the UM Offense is slightly below average, while the MSU Defense is well above average, pulling the UM Offense down to well below average. On the other side, the MSU Offense is below average, but the UM Defense is above average, which pushes the MSU Offense downward. The net matchup, however, is a sizeable Field Position edge for MSU.

Finishing Drives

In Finishing Drives, both offenses are below average, and both defenses are above average. The key difference is UM’s Defense, however, which is well above average. The net matchup is a considerable advantage for Michigan in Finishing Drives.

Turnover Luck

Both teams have a recent history of having poor Turnover Luck. The story of the season at this point is that both MSU and Michigan’s TO Luck has lagged expectation based on measurables (e.g. Fumbles and Passes Defended). As much as MSU likes to complain about its lack of ability to create turnovers, the opportunities it’s had for effecting turnovers pales in comparison to Michigan’s. In the end, Michigan’s TO Luck has been significantly worse than MSU’s, to the tune of about 5.2 PPG. The net matchup is a significant advantage for Michigan in TO Luck.

Rushing Matchups

Not to belabor each matchup as much as above, but here Sparty appears to have a net advantage in all 5 of the Rushing matchups, and most by a sizeable amount. The issues that the Wolverines are having with its Offensive Line become apparent when looking at these characteristics. From the looks of it, the Michigan rushing attack is heading for some bloody tough sledging, as they say on the other side of the pond.

Rushing Success Rate

In Rushing Success, both offenses are below average, while both defenses are above average. The difference is that UM’s Offense is well below average (#103), while Sparty’s Defense is elite (#4). The net matchup balance is a sizeable advantage in Rushing Success for Sparty.

Explosiveness

In Explosiveness, the UM Offense is well above average, but the MSU Defense is also above average, pulling the UM Offense down close to average. On the other side, the MSU Offense is above average, but the UM Defense is perfectly average. In the end, Rushing IsoPPP (Explosiveness) favors Sparty by the thinnest of margins.

Opportunity Rate

In Opportunity Rate, the UM Offense is well below average, and the MSU Defense is above average, pushing the UM Offense down to a woeful level. On the other side, the MSU Offense is below average, and the UM Defense is above average. The net is a considerable advantage for Sparty in Opportunity Rate.

Power Success Rate

In Power Success Rate, both defenses are well above average: Michigan is ranked #7, while MSU is elite, ranked #2. The UM Offense is above average, while the MSU Defense is below average. However, the MSU Defense is so good in this category, the matchup balance is a sizeable advantage for MSU in Power Success Rate.

Stuff Rate

Last is Stuff Rate (a contra-metric). In this case, once again, both offenses are below average, while the defenses are both above average. In the end, the matchup result is a sizeable advantage for MSU in Stuff Rate.

Passing Matchups

The Passing matchups are split.

Passing Success Rate

In Passing Success Rate, the UM Offense is below average, and when matched against the well above average MSU Defense, it pulls the UM Offense even further down. On the other side, the well above average MSU Offense is obliterated by the elite UM Defense (ranked #1 in this category). The net is a significant advantage for Michigan in Passing Success Rate.

Passing Explosiveness

In Passing IsoPPP (Explosiveness), the above average UM Offense is negated by the MSU Defense. On the other side, the well below average MSU Offense is also similarly negated by the well below average UM Defense. The matchup result is a negligible advantage for MSU in Passing IsoPPP.

Standard Down Matchups

Michigan State captures 3 of the 4 Standard Down matchups with Michigan, but UM’s advantage in SD Line Yards nearly offsets MSU’s only significant advantage in SD IsoPPP. It’s worth noting that UM’s defensive scheme under Harbaugh, and under Don Brown in particular, is usually weak in the Explosiveness metric. However, it is usually offset by a strong Success Rate metric, which means that although the explosive plays given up may tend to be large, they are a very infrequent.

SD Success Rate

In SD Success Rate, both defensive units are elite: MSU is #4 and UM is #2. Meanwhile, the MSU Offense is below average, and the UM Offense is extremely below average (#120). The net matchup result is a marginal advantage for MSU in SD Success Rate.

SD Explosiveness

In SD Explosiveness, the UM Offense is about average, but the MSU Defense is elite (#3), which pushes the UM Offense far downward. On the other side, the MSU Offense is below average, but the UM Defense is about equally below average, making the MSU Offense look about average. The net matchup result is a significant advantage for MSU in SD Explosiveness.

SD Line Yards per Carry

In SD LYPC, the UM Offense is significantly below average, and the MSU Defense is significantly above average, pulling the UM Offense even further downward. However, the MSU Offense is below average, while the UM Defense is elite (ranked #4) and pushes the MSU Offense down to an even lower level. The net matchup result is a significant advantage for Michigan in SD Line Yards per Carry.

SD Sack Rate

In SD Sack Rate (a contra-metric), the UM Offense is significantly below average, while the MSU Defense is above average, which pushes the UM Offense up. On the other side, the MSU Offense is similarly above average, but the UM Defense is elite (ranked #1), which pulls the MSU Offense up as well. In the end, the net matchup result is a slight advantage for MSU in SD Sack Rate.

Passing Down Matchups

Last, but certainly not least, are the Passing Down matchups, which show three out of four metrics tilting toward Michigan’s advantage. The bottom line is, an opponent like Michigan State does not want to be in a passing situation against Michigan. Also, the Michigan offense may have a much better day in passing situations against this Michigan State defense, as long as it can manage to avoid the pass rush by sticking to shorter passes to the slot or one of multiple TE’s, and working the play action regimen thoroughly.

PD Success Rate

In PD Success Rate, the UM Offense is above average while the MSU Defense and its sorry safeties are below average, which tweaks the UM Offense up slightly. On the other side, the top 10 MSU Offense is still obliterated by the elite UM Defense (ranked #2). The net matchup result is a significant advantage for Michigan in PD Success Rate.

PD Explosiveness

In PD Explosiveness (IsoPPP), here again the UM Offense is well above average, while the MSU Defense is actually above average and pulls the UM Offense down a bit. On the other side, the MSU Offense is below average, and the UM Defense is about average, which tweaks the MSU Offense up a little. The net matchup result is still a sizeable advantage for Michigan in PD Explosiveness.

PD Line Yards per Carry

In PD Line Yards per Carry (LYPC), the UM Offense is above average and the MSU Defense is below average, which boosts the UM Offense up a smidge. On the other side, the MSU Offense is well above average, but the UM Defense is also above average, which takes the MSU Offense down a notch. The net matchup result is a slight advantage for Michigan in PD LYPC.

PD Sack Rate

In PD Sack Rate (a contra-metric), as everyone should know by now, the UM Offense is well below average, while the MSU Defense is equally above average, which boosts the UM Offense even higher. On the other side, the MSU Offense is below average and the UM Defense is well above average, which also kicks the MSU Offense up a good bit. Yet, the net matchup result is a sizeable advantage for MSU in PD Sack Rate.

Conclusion

So at this point you may have some mixed feelings about this mixed back of metrics matchups. Your gut is all a-flutter, and visions of Blake O-Neill muffing a long snap are corrupting your visions of grandeur and the magic that is Under the Lights. You're looking for the Tums. It's time to take a long, deep breath. Inhale. Now count to 10 while you exhale. Now do that 10 times. You better now? Good.

Dennis Hopper - Blue Velvet

Now first, just remember that the core Five Factors are significantly to Michigan's advantage here. Michigan has an 83% likelihood to win this game. That's 1:5 odds. That means you need to bet $5 to win $1 from you Sparty friend. What's not included in that margin is Turnover Luck, which of course is random. But if you believe in the principle of Regression to the Mean, Michigan has been consistently absorbing nearly 7 points per game of bad Turnover Luck. Which means, if Michigan turns it over, it has the ability to make up for it by creating opportunities to get the ball back, or just playing that much better otherwise. We've seen that, and that's what the stats tell us so far this year. If Sparty turns the ball over, they're dead meat.

This game will be much like the Purdue game. If Michigan can just get a lead, it's over. This defense is not going give it up. Now the question is then, how can Michigan get - or take - the lead? The key is going to be a balanced attack, which means Michigan is going to need to actually throw the ball more than it ever has this season. About 25% more, in fact.

Thus far this season, M is running the ball on over 60% of all its plays. Once M gets ahead by more than a TD, the rush attempt rate goes up to 70%, and 85% when ahead by more than 14. OK, that's fine, but...when M is leading by less than a score, tied or trailing, it's still running on 55-60% of downs, but it's only making 35% of its yardage then as YPC drops from 6.1 (when ahead by 14) to 3.5 (when behind by 7). Now imagine how things might go against the Sparty front seven given the metrics above. Yet, on the passing side, YPA is consistently in the 7.0 to 10.0 range whether ahead or behind, which is not terrible, actually.

So what's needed is somewhere around 35 pass attempts, using quick passes to the slot or one of the mutiple TE's on slants or a mesh. Sprinkle in an occasional play action to get O'Korn free of the rush on a waggle, say. O'Korn is Michigan's X-factor here. That's the key: O'Korn needs to be kept clean, and it can't be assumed a pocket is going to exist for very long. No seven- or even five-step drops. He's highly mobile, steps quickly through his progressions and can throw well on run. Avoiding the rush and getting the passes off will compel MSU to back out of the box. Once that happens, it's time to eat the MSU safeties alive with McKeon and Gentry running skinny posts until the cows come home.

So, that concludes this week’s Five Factors Matchup Analysis!

Yours in football, and Go Blue!

Comments

TrueBlue2003

October 6th, 2017 at 1:46 AM ^

but the guy calls State's rushing success rate against elite at 30.0% (4th) but ours is 30.4% (7th).

And their offense is 93rd in the country, whereas ours is 103rd.  Not that big of a difference and that's somehow a sizable advantage for them?  It's pretty much negligible.

Also, these aren't opponent adjusted rates.  When you play Air Force, that'll bring down your raw rushing defensive success rates.

I would take our rushing offense against their D easily.

ish

October 5th, 2017 at 12:47 PM ^

is there a conclusion?  i can see the outcome of each of the five factors, but don't quite get how they relate sufficiently to know how i should feel.  please provide feelings!

TESOE

October 5th, 2017 at 1:25 PM ^

this is going to be a good game.  Throw the records out.  Throw the advanced metrics in.

I'm surprised  at the Spartan strengths.  This doesn't match my gut nor my box score record keeping.

I am ready and excited.  Hail Yes! ... indeed.

Thanks.

Prince_of_Nachos

October 5th, 2017 at 4:16 PM ^

Really excellent data visualization. Summarizes a lot of information in a well laid out, readable format.

Would be a great addition to the game preview each week.

Side note - what's up with the light gray horizontal bars on the first chart? Almost looks like they're logarithmically spaced. 

J.

October 5th, 2017 at 9:57 PM ^

Sorry, but your analysis of agression to the mean appears to contain a common fallacy. Michigan does not have an advantage in turnover luck due to regression to the mean. If turnovers are random, neither team has an advantage — full stop. Your suggestion proposes that Michigan is “due” to have good turnover luck because it’s had bad luck in the past, but that’s not the way luck works. Over the long term, you could expect everyone’s luck to be 0, but that needn’t be true over any finite amount of time.

The rest of the analysis seems good, but I think you should just leave that one out. Perhaps % of chaos plays on offense and defense might be a good substitute? Perhaps some teams are better than others at creating (or limiting) the situations that are likely to lead to turnovers. But the actual luck — which way the ball bounces — isn’t predictive.

Ecky Pting

October 6th, 2017 at 12:21 AM ^

I fundamentally agree with you, so I'm having another look at this. The reason Michigan might expect better turnover results in its next game is not because it's had bad turnover luck and they're due... It's because to date, it's measurable statistics that actually correlate with turnovers, namely fumbles and passes defended, have far exceeded the actual results. The intent is to capture the expected turnovers. Typically fumbles split about 50/50, and a pass defended scales to about 0.2-0.25 interceptions. The expected turnovers metric aggregates these stats and is expressed as an expected number of points attributable to turnovers. Michigan is ranked #4 (+5.3 ppg). In actual turnovers generated, M ranks only #63 (0 ppg). The difference between expected and actual values is the "turnover luck".

One can never predict the luck component, of course, but if one were to attempt to predict the points off of turnovers based on past statistics, the expected number would be applicable. So in revisiting this, taking the difference between the turnover luck factors I agree is not entirely correct. It should really be the difference in the expected values only. The actual values shouldn't enter into the prediction since they would in effect be double-counting the predictors, as well as including the luck factor.

Looking at MSU, they're ranked #112 (-2.8 ppg) in expected points off turnovers, while their actual results are ranked #104 (-4.0 ppg). So despite their lamentations in the media about being plagued by turnovers, they're actually doing better than they should reasonable expect in terms of their ranking at least, but I digress...

Taking the difference in expected points between M & MSU, M would still be at an expected advantage of about 8.1 ppg (5.3 - (-2.8)).

Magnum P.I.

October 6th, 2017 at 10:13 AM ^

We'd do ourselves an emotional favor by treating this game like a toss-up. 

We'll hold State to about 14-24 points, because you know they'll break a few big ones against us--they always, always do.

They will stymie our run game, and most of the game while we're on offense will be incredibly frustrating. I wouldn't count on more than 100 rushing yards as a team.

It reallly hinges on O'Korn, who is a mostly unknown quantity, and the offensive play-calling and whether it's creative enough to exploit State's weak areas. There should be no question what State's gameplan will be: stack the box; stop the run; pressure the QB. Can we take advantage enough to get the 20-30 points we'll likely need?